Home >Documents >EO4SD-Urban DAKAR - City Operations Report · EO4SD-Urban Project: Dakar City Report Partners:...

EO4SD-Urban DAKAR - City Operations Report · EO4SD-Urban Project: Dakar City Report Partners:...

Date post:01-Jun-2020
Category:
View:1 times
Download:0 times
Share this document with a friend
Transcript:
  • Earth Observation for Sustainable Development

    Urban Development Project

    This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 685761.

    ESA Ref: AO/1-8346/15/I-NB

    Doc. No.: City Operations Report

    Issue/Rev.: 1.1

    Date: 19.11.2019

    EO4SD-Urban Project: Dakar City Report

    Partners: Financed by:

  • Earth Observation for Sustainable Doc. No.: City-Operations Report

    Development – Urban Project Issue/Rev-No.: 1.1

    EO4SD-Urban Dakar City Operations Report Page I

    Consortium Partners

    No. Name Short Name Country

    1 GAF AG GAF Germany

    2 Système d'Information à Référence Spatiale SAS SIRS France

    3 GISAT S.R.O. GISAT Czech Republic

    4 Egis SA EGIS France

    5 Deutsche Luft- und Raumfahrt e. V DLR Germany

    6 Netherlands Geomatics & Earth Observation B.V. NEO The Netherlands

    7 JOANNEUM Research Forschungsgesellschaft mbH JR Austria

    8 GISBOX SRL GISBOX Romania

    Disclaimer:

    The contents of this document are the copyright of GAF AG and Partners. It is released by GAF AG

    on the condition that it will not be copied in whole, in section or otherwise reproduced (whether by

    photographic, reprographic or any other method) and that the contents thereof shall not be divulged to

    any other person other than of the addressed (save to the other authorised officers of their organisation

    having a need to know such contents, for the purpose of which disclosure is made by GAF AG)

    without prior consent of GAF AG.

  • Earth Observation for Sustainable Doc. No.: City-Operations Report

    Development – Urban Project Issue/Rev-No.: 1.1

    EO4SD-Urban Dakar City Operations Report Page II

    Summary

    This document contains information related to the provision of geo-spatial products from the European

    Space Agency (ESA) supported project “Earth Observation for Sustainable Development” Urban

    Applications (EO4SD-Urban) to the benefit of Global Platform for Sustainable Cities (GPSC)

    programme implemented for the City of Dakar and Senegalese authorities.

    Affiliation/Function Name Date

    Prepared SIRS

    NEO

    JR

    S. Delbour, D. Fretin,

    V. Gastal

    F. Fang

    M. Hirschmugl, H.

    Proske

    26/09/2019

    Reviewed SIRS C. Sannier 27/09/2019

    Approved GAF AG, Project Coordinator T. Haeusler 02/10/2019

    The document is accepted under the assumption that all verification activities were carried out

    correctly and any discrepancies are documented properly.

    Distribution

    Affiliation Name Copies

    ESA Z. Bartalis electronic copy

    Government agencies M. Ndaw

    F. Cheikh

    M. Diara

    electronic copy

    UNIDO M. Draeck

    K. Barunica

    electronic copy

    Document Status Sheet

    Issue Date Details

    1.0 02/10/2019 First Document Issue

    1.1 19/11/2019 Addition of section 4.4 Sustainable Development Goal 11 Indicators

    Document Change Record

    # Date Request Location Details

    1 02/10/2019 Initial version

  • Earth Observation for Sustainable Doc. No.: City-Operations Report

    Development – Urban Project Issue/Rev-No.: 1.1

    EO4SD-Urban Dakar City Operations Report Page III

    This page is intentionally left blank.

  • Earth Observation for Sustainable Doc. No.: City-Operations Report

    Development – Urban Project Issue/Rev-No.: 1.1

    EO4SD-Urban Dakar City Operations Report Page IV

    Executive Summary

    The European Space Agency (ESA) has been working closely together with the International Finance

    Institutes (IFIs) and their client countries to demonstrate the benefits of Earth Observation (EO) in the

    IFI development programmes. Earth Observation for Sustainable Development (EO4SD) is a new

    ESA initiative, which aims to achieve an increase in the uptake of satellite-based information in the

    regional and global IFI programmes. The overall aim of the EO4SD Urban project is to integrate the

    application of satellite data for urban development programmes being implemented by the IFIs or

    Multi-Lateral Development Banks (MDBs) with the developing countries. The overall goal will be

    achieved via implementation of the following main objectives:

    • To provide a service portfolio of Baseline and Derived urban-related geo-spatial products

    • To provide the geo-spatial products and services on a geographical regional basis

    • To ensure that the products and services are user-driven

    This Report describes the generation and the provision of EO-based information products to the GEF

    supported programme “Global Platform for Sustainable Cities” for Senegal and the counterpart City

    authorities in Dakar. The Report provides a Service Description by referring to the user-driven service

    requirements and the associated product list with the detailed product specifications. The following

    products were requested:

    • Urban and Peri-Urban Land Use/ Land Cover and Changes

    • Settlement Extent and Imperviousness and Changes

    • Urban Green Areas and Changes

    • Flood Hazard and Risk Assessment

    The current Version of this Report contains the description of the generation and delivery of each

    requested product, especially the Land Use/Land Cover (LU/LC) and the LU/LC Changes between

    2006 and 2018. The Urban Green Areas and Flood Hazard and Risk Assessment study conducted, and

    the resulting products are also described in detail in this Report.

    This City Operations Report for Dakar systematically reviews the main production steps involved and

    importantly highlights the Quality Control (QC) mechanisms involved; the steps of QC and the

    assessment of quality is provided in related QC forms in the Annexe of this Report. Standard

    analytical work undertaken with the products can be further included as inputs into further urban

    development assessments, modelling and reports.

  • Earth Observation for Sustainable Doc. No.: City-Operations Report

    Development – Urban Project Issue/Rev-No.: 1.1

    EO4SD-Urban Dakar City Operations Report Page V

    This page is intentionally left blank.

  • Earth Observation for Sustainable Doc. No.: City-Operations Report

    Development – Urban Project Issue/Rev-No.: 1.1

    EO4SD-Urban Dakar City Operations Report Page VI

    Table of Contents

    1 GENERAL BACKGROUND OF EO4SD-URBAN ................................................................... 1 2 SERVICE DESCRIPTION........................................................................................................... 1

    2.1 STAKEHOLDERS AND REQUIREMENTS ..................................................................................... 1

    2.2 SERVICE AREA SPECIFICATION ............................................................................................... 2

    2.3 PRODUCT LIST AND PRODUCT SPECIFICATIONS ...................................................................... 5

    2.4 LAND USE/LAND COVER NOMENCLATURE ............................................................................. 5

    2.5 WORLD SETTLEMENT EXTENT ................................................................................................ 9

    2.6 PERCENTAGE IMPERVIOUS SURFACE .................................................................................... 10

    2.7 URBAN GREEN AREAS NOMENCLATURE .............................................................................. 10

    2.8 TERMS OF ACCESS ................................................................................................................. 10

    3 SERVICE OPERATIONS .......................................................................................................... 11

    3.1 SOURCE DATA ....................................................................................................................... 11

    3.2 PROCESSING METHODS ......................................................................................................... 12

    3.3 ACCURACY ASSESSMENT OF MAP PRODUCTS ...................................................................... 12

    3.3.1 Accuracy Assessment of the LU/LC Products ................................................................................ 12

    3.3.2 Accuracy Assessment of the World Settlement Extent Product ...................................................... 18

    3.3.3 Accuracy Assessment of the Percentage Impervious Surface Product ............................................ 20

    3.3.4 Accuracy Assessment of Urban Green Areas Product .................................................................... 22

    3.3.5 Accuracy Assessment of Flood Extent Product .............................................................................. 23

    3.4 QUALITY CONTROL/ASSURANCE .......................................................................................... 25

    3.5 METADATA ............................................................................................................................ 26

    4 ANALYSIS OF MAPPING RESULTS ..................................................................................... 27

    4.1 SETTLEMENT EXTENT – DEVELOPMENTS 2000, 2005, 2010 AND 2015 ................................ 27

    4.2 LAND USE / LAND COVER 2003/2006 AND 2018 ................................................................... 29

    4.2.1 LU/LC Mapping for Core City Area ............................................................................................... 29

    4.2.2 Spatial Distribution of Main LU/LC Change Categories for Core City Area ................................. 33

    4.2.3 LU/LC Mapping for Larger Urban Area ......................................................................................... 35

    4.2.4 Spatial Distribution of Main LU/LC Change Categories for Larger Urban Area ........................... 37

    4.3 URBAN GREEN AREAS ........................................................................................................... 40

    4.4 SUSTAINABLE DEVELOPMENT GOAL 11 INDICATORS ........................................................... 42

    4.4.1 SDG 11 Indicator 11.2.1 ................................................................................................................. 43

    4.4.2 SDG 11 Indicator 11.3.1 ................................................................................................................. 44

    4.4.3 SDG 11 Indicator 11.7.1 ................................................................................................................. 46

    4.5 CONCLUDING POINTS ............................................................................................................ 47

    5 FLOOD HAZARD AND RISK ASSESSMENT ....................................................................... 48

    5.1 GENERAL CHARACTERISTICS OF THE STUDY AREA .............................................................. 49

  • Earth Observation for Sustainable Doc. No.: City-Operations Report

    Development – Urban Project Issue/Rev-No.: 1.1

    EO4SD-Urban Dakar City Operations Report Page VII

    5.2 FLOOD HISTORY .................................................................................................................... 58

    5.3 EO DATA USED ..................................................................................................................... 62

    5.4 SHORT DESCRIPTION OF METHODOLOGICAL APPROACH ....................................................... 63

    5.5 PRODUCT DESCRIPTION AND ACCURACY ASSESSMENT ....................................................... 67

    5.6 RESULTS ................................................................................................................................ 69

    6 REFERENCES ............................................................................................................................ 75

    Annexes

    Annex 1: AOI calculation based on the DG Regio approach

    Annex 2: Processing methods for EO4SD-Urban products

    Annex 3: Filled Quality Control Sheets

    List of Figures

    Figure 1: Illustration of Core City and Larger Urban Areas of Dakar. ................................................... 4 Figure 2: Mapping result of the Core City Area of Dakar of the year 2018 overlaid with randomly

    distributed sample points used for accuracy assessment. ...................................................................... 15 Figure 3: Mapping result of the Larger Urban Area of Dakar of the year 2018 overlaid with randomly

    distributed sample points used for accuracy assessment. ...................................................................... 16 Figure 4: Result of the Urban Green Area mapping in Dakar (change product) with sampling points

    used for product validation. ................................................................................................................... 22 Figure 5: Result of the Flood extent mapping in Dakar with sampling points used for product

    validation. .............................................................................................................................................. 24 Figure 6: Quality Control process for EO4SD-Urban product generation. At each intermediate

    processing step output properties are compared against pre-defined requirements. ............................. 25 Figure 7: Settlement Extent developments in the epochs 2000 to 2005, 2005 to 2010 and 2010 to 2015

    in Dakar and surrounding region. .......................................................................................................... 27 Figure 8: Settlement Extent developments in the epochs 2000 to 2005, 2005 to 2010 and 2010 to 2015

    in Dakar within the High Density Area. ................................................................................................ 28 Figure 9: Core City Area - Detailed LU/LC 2018 in Dakar .................................................................. 29 Figure 10: Core City Area - Insight on the detailed Land Use Land Cover 2018 inside the city.......... 30 Figure 11: Core City Area - Detailed LU/LC 2006 structure, in % (left) and km2 (right). ................... 31 Figure 12: Core City Area - Detailed LU/LC 2018 structure, in % (left) and km2 (right). ................... 31 Figure 13: Core City Area – LU/LC change types and spatial distribution .......................................... 33 Figure 14: Core City Area – LU/LC Change types between 2006 and 2018 presented in % (left) and

    sqkm (right). .......................................................................................................................................... 34 Figure 15: Larger Urban area – LU/LC 2018 in Dakar. ........................................................................ 35 Figure 16: Larger Urban Area - Insight on the Land Use Land Cover 2018 on the urban fringe. ........ 35 Figure 17: Larger Urban Area - Detailed LU/LC 2006 structure presented in % (left) and km2 (right).

    ............................................................................................................................................................... 36 Figure 18: Larger Urban Area - Detailed LU/LC 2018 structure presented in % (left) and km2 (right).

    ............................................................................................................................................................... 36 Figure 19: Larger Urban Area – LU/LC Change types and spatial distribution. .................................. 38 Figure 20: Larger Urban Area – LU/LC Change types 2006 -2018 area in % (left) and sqkm (right) . 38 Figure 21: Urban Green Areas changes and spatial distribution. .......................................................... 40 Figure 22: Status and change of urban green areas in-between 2004/2006 and 2018 expressed in %. . 41 Figure 23: Status and change of urban green areas in-between 2004/2006 and 2018 expressed in area.

    ............................................................................................................................................................... 41

  • Earth Observation for Sustainable Doc. No.: City-Operations Report

    Development – Urban Project Issue/Rev-No.: 1.1

    EO4SD-Urban Dakar City Operations Report Page VIII

    Figure 24: Proportion of population with convenient access to public transport. ................................. 43 Figure 25: Ratio of land consumption rate to population growth rate between 2005 and 2015. ........... 44 Figure 26: Percentage change of population and land consumption between 2005 and 2015. ............. 45 Figure 27: Average share of the built-up area that is open space for public use. .................................. 47 Figure 28: Four days after a storm in August 2015, flood waters are still visible in N'Gor Village,

    Dakar, Senegal (Photo: Jürgen Fauth, BRACED) ................................................................................. 48 Figure 29: Position of main parts of Dakar (taken from Wikipedia) ..................................................... 49 Figure 30: Dakar, Senegal – Service Area: pink: Core City Area of Interest; green: Larger Urban Area

    of Interest (Background Image: Sentinel 2, recorded on 10/10/2016, European Space Agency) ......... 50 Figure 31: Altitudes in Dakar Service Area as derived from available Digital Terrain and Surface

    Models (western part: 5m Digital Terrain Model of Dakar (BaseGéo Sénégal,

    (http://www.basegeo.gouv.sn/) based on Urban Database (UDB) product; eastern part: ALOS Global

    Digital Surface Model "ALOS World 3D - 30m (AW3D30)", version 2.1 (©JAXA): Amthyst areas

    indicate most flood prone zones (altitudes below 5 m). ........................................................................ 51 Figure 32: Climate average from 2000 to 2012 in Dakar, Source: World Weather Online .................. 52 Figure 33: Tally Neitty Mbar (main street of Djeddah Thiaroye Kao in the Department of Pikine),

    flooded in 2009. Source: Requalification des zones inondés de Djeddah Thiaroye Kao.urbaDTK.org 54 Figure 34: Possible Flooding Area in the New Urban Expansion Area based on Flo2D modelling

    results (taken from JICA 2016) ............................................................................................................. 55 Figure 35: A pump to clear out flooded streets of the Pikine neighbourhood of the Senegal capital

    Dakar PROGEP measures (Photo Mamadou Lamine Camara, Agence de développement municipal

    (ADM) Dakar) ....................................................................................................................................... 57 Figure 36: In the "streets" of Dakar, 17/09/2009 - Photo: SOS Archives ............................................. 59 Figure 37: Flooded settlement in the Department of Pikine, August/September 2012 (Photo: Steve

    Cockburn) .............................................................................................................................................. 60 Figure 38: Flooding in Pikine, Sptember 2012 (Senegal7.com) ........................................................... 60 Figure 39: Coverage of the 5m Digital Terrain Model of Dakar (© BaseGéo Sénégal) ....................... 63 Figure 40: Flooded areas in southern part of Pikine (neighbourhood of Diammaguen) in August 2015

    (image recorded on 27/08/2015, © Maxar Technologies)..................................................................... 64 Figure 41: Flooded areas in southern part of Pikine (neighbourhood of Dalifort) in August 2017

    (image recorded on 13/08/2017, © Maxar Technologies)..................................................................... 64 Figure 42: Subset of Flood Hazard Map of Dakar (Department of Pikine) (Background Image:

    Sentinel 2, recorded on 10/10/2016, European Space Agency) ............................................................ 68 Figure 43: Subset of Flood Risk Map of Dakar (Department of Pikine) (Background Image: Sentinel 2,

    recorded on 10/10/2016, European Space Agency) .............................................................................. 68 Figure 44: Percentages of flood hazard zones in Dakar core city area .................................................. 69 Figure 45: Percentages of flood hazard zones in Dakar peri-urban region ........................................... 70 Figure 46: Proportion of Residential Urban Fabric in flood hazard zones in Dakar core city area ...... 70 Figure 47: Subset of map of Residential, Industrial, Commercial and Public Urban Fabric combined

    with Flood Hazard Zoning in Dakar’s department of Pikine (Background Image: Sentinel 2, recorded

    on 10/10/2016, European Space Agency) ............................................................................................. 71 Figure 48: Percentages of flood risk zones in Dakar Core city Area .................................................... 72 Figure 49: Percentages of flood risk zones in Dakar Larger Urban Area ............................................. 72 Figure 50: Subset of map of Residential, Industrial, Commercial and Public Urban Fabric combined

    with Flood Risk Zoning in Dakar’s department of Pikine (Background Image: Sentinel 2, recorded on

    10/10/2016, European Space Agency) .................................................................................................. 73 Figure 51: Proportion of Residential Urban Fabric in flood hazard zones in Dakar Core City Area .... 74 Figure 48: Satellite image showing Melaka and the surrounding area. ................................................. 81 Figure 49: Global Human Settlement Population Layer (spatial resolution of 1 km). .......................... 81 Figure 50: DLR population layer (spatial resolution of 10 m). ............................................................. 81 Figure 51: Aggregated DLR population layer (spatial resolution of 1 km). ......................................... 82

  • Earth Observation for Sustainable Doc. No.: City-Operations Report

    Development – Urban Project Issue/Rev-No.: 1.1

    EO4SD-Urban Dakar City Operations Report Page IX

    Figure 52: High Density Core area of Melaka calculated based on the aggregated DLR population

    layer. The image on the left shows the AOI overlaid on the DLR population layer. On the right, the

    AOI is overlaid on a RGB satellite image. ............................................................................................ 82 Figure 53: Urban Cluster area of Melaka calculated based on the aggregated DLR population layer.. 83

    List of Tables

    Table 1: LU/LC Nomenclature for GPSC Cities (Core City AOI). ........................................................ 7 Table 2: LU/LC Nomenclature for GPSC Cities (Larger Urban AOI). .................................................. 7 Table 3: Number of sampling points for the Core City Area classes after applied sampling design with

    information on overall land cover by class. ........................................................................................... 14 Table 4: Number of sampling points for the Larger Urban Area classes after applied sampling design

    with information on overall land cover by class. .................................................................................. 14 Table 5: Accuracies exhibited by the WSF2015 according to the three considered agreement criteria

    for different definitions of settlement. ................................................................................................... 19 Table 6: Acquisition dates and size of the WV2 images available for the 5 test sites analysed in the

    validation exercise along with the number of corresponding 30x30m validation samples. .................. 21 Table 7: Results of the accuracy assessment of flood extents in Dakar - Overall Accuracy 90.33 %. . 24 Table 8: Detailed information on area and percentage of total area for each class for 2006 and 2018 as

    well as the changes for the Core City area ............................................................................................ 32 Table 9: Overall Main LU/LC Changes Statistics for the Core City Area. ........................................... 34 Table 10: Larger Urban Area - Detailed information on area and percentage of total area for each class

    for 2006 and 2018 as well as the changes. ............................................................................................ 37 Table 11: Overall LU/LC statistics of the Larger Urban Area. ............................................................. 39 Table 12: SDG 11 indicators measurable with the support of EO4SD-Urban products. ...................... 42 Table 13: Land use classes and reclassification to pre-defined damage levels in Core City Area ........ 66 Table 14: Land use classes and reclassification to pre-defined damage levels in Larger Urban Area .. 66 Table 15: Flood Hazard and Risk classification .................................................................................... 67

  • Earth Observation for Sustainable Doc. No.: City-Operations Report

    Development – Urban Project Issue/Rev-No.: 1.1

    EO4SD-Urban Dakar City Operations Report Page X

    List of Abbreviations

    AOI Area of Interest

    CDS City Development Strategy

    CS Client States

    DEM Digital Elevation Model

    DLR German Space Agency

    EDF European Development Fund

    EEA European Environmental Agency

    EGIS Consulting Company for Environmental Impact Assessment and Urban Planning, France

    EO Earth Observation

    ESA European Space Agency

    EU European Union

    GAF GAF AG, Geospatial Service Provider, Germany

    GCC General Clauses and Conditions for ESA Contracts

    GCT General Conditions of Tender

    GEO Group on Earth Observations

    Geo-SDI Geo Sustainable Development Indicators

    GIS Geographic Information System

    GISAT Geospatial Service Provider, Czech Republic

    GISBOX Romanian company with activities of Photogrammetry and GIS

    GPSC Global Platform for Sustainable Cities

    GUF Global Urban Footprint

    HR High Resolution

    HRL High Resolution Layer

    IFI International Financing Institute

    INSPIRE Infrastructure for Spatial Information in the European Community

    ISO/TC 211 Standardization of Digital Geographic Information

    JR JOANNEUM Research, Austria

    LU / LC Land Use / Land Cover

    LULCC Land Use and Land Cover Change

    MMU Minimum Mapping Unit

    NDVI Normalized Difference Vegetation Index

    NEO Geospatial Service Provider, The Netherlands

    QA Quality Assurance

    QC Quality Control

    QM Quality Management

    R&D Research and Development

    SAR Synthetic Aperture Radar

    SC Service Cluster

    SIRS Geospatial Service Provider, France

    SME Small and Medium-sized Enterprise

    SO Service Operations

  • Earth Observation for Sustainable Doc. No.: City-Operations Report

    Development – Urban Project Issue/Rev-No.: 1.1

    EO4SD-Urban Dakar City Operations Report Page XI

    SP Service Provider

    ToC Table of Contents

    UN United Nations

    UNDP United Nations Development Programme

    UN-ESCAP United Nations Economic and Social Commission for Asia and the Pacific

    UNFCCC United Nations Framework Convention on Climate Change

    UNIDO United Nations Industrial Development Organisation

    UNITAR United Nations Institute for Training and Research

    US United States of America

    UUA User Utility Assessment

    VHR Very High Resolution

    WB World Bank

    WBG World Bank Group

    WP Work Package

  • Earth Observation for Sustainable Doc. No.: City-Operations Report

    Development – Urban Project Issue/Rev-No.: 1.1

    EO4SD-Urban Dakar City Operations Report Page XII

    This page is intentionally left blank.

  • Earth Observation for Sustainable Doc. No.: City-Operations Report

    Development – Urban Project Issue/Rev-No.: 1.1

    EO4SD-Urban Dakar City Operations Report Page 1

    1 General Background of EO4SD-Urban

    Since 2008 the European Space Agency (ESA) has worked closely together with the International

    Finance Institutes (IFIs) and their client countries to harness the benefits of Earth Observation (EO) in

    their operations and resources management. Earth Observation for Sustainable Development (EO4SD)

    is a new ESA initiative, which aims to achieve an increase in the uptake of satellite-based information

    in the regional and global IFI programmes. The EO4SD-Urban project initiated in May 2016 (with a

    duration of 3 years) has the overall aim to integrate the application of satellite data for urban

    development programmes being implemented by the IFIs with the developing countries. The overall

    goal will be achieved via implementation of the following main objectives:

    • To provide the services on a regional basis (i.e. large geographical areas); in the context of the

    current proposal with a focus on S. Asia, SE Asia and Africa, for at least 35-40 cities.

    • To ensure that the products and services are user-driven; i.e. priority products and services to

    be agreed on with the MDBs in relation to their regional programs and furthermore to

    implement the project with a strong stakeholder engagement especially in context with the

    validation of the products/services on their utility.

    • To provide a service portfolio of Baseline and Derived urban-related geo-spatial products that

    have clear technical specifications and are produced on an operational manner that are

    stringently quality controlled and validated by the user community.

    • To provide a technology transfer component in the project via capacity building exercises in

    the different regions in close co-operation with the MDB programmes.

    This Report supports the fulfilment of the third objective, which requires the provision of geo-spatial

    Baseline and Derived geo-spatial products to various stakeholders in the IFIs and counterpart City

    authorities. The Report provides a Service Description, and then in Chapter 3 systematically reviews

    the main production steps involved and importantly highlights whenever there are Quality Control

    (QC) mechanisms involved with the related QC forms in the Annexe of this Report. The description of

    the processes is kept intentionally at a top leave and avoiding technical details as the Report is

    considered mainly for non-technical IFI staff and experts and City authorities. Finally, Chapter 4

    presents the standard analytical work undertaken with the products which can be inputs into further

    urban development assessments, modelling and reports.

    2 Service Description

    The following Section summarises the service as it has been realised for the city of Dakar, Senegal,

    within the EO4SD-Urban Project and as it was delivered to the UNIDO (United Nations Industrial

    Development Organisation), the GPSC Implementing Agency for the Senegalese city of Dakar, and

    the Senegalese Governmental agencies.

    2.1 Stakeholders and Requirements

    The EO4SD-Urban products described in this Report were provided for the benefit of the Global

    Platform for Sustainable Cities (GPSC) programme. GPSC is funded by the Global Environment

    Facility (GEF) and currently includes 28 cities in 11 countries. The GPSC initiative is supported by

    different Multi-Lateral Development Banks (MDBs) and UN organisations. The two Senegalese cities

    of St. Louis and Dakar are part of the GPSC programme and were identified for collaboration with the

    EO4SD-Urban project via interactions with UNIDO, the Implementing Agency.

  • Earth Observation for Sustainable Doc. No.: City-Operations Report

    Development – Urban Project Issue/Rev-No.: 1.1

    EO4SD-Urban Dakar City Operations Report Page 2

    The GPSC has an overarching aim to provide a knowledge platform for partner cities, as well as

    relevant networks and institutions to support the cities via:

    • “Knowledge transfer activities that support urban investments and sustainability initiatives,

    • A global network for collaborative engagement, tapping into and complementing existing

    efforts,

    • Long-term, systematic engagement with cities, financial institutions, and organizations for

    transformational impact” (GPSC Programme Booklet, 2016).

    Dakar is a GPSC Partner City and the capital city of Senegal. Located along the Atlantic coast in the

    east of the country, this geographical position necessarily exposes the city to major environmental

    hazards (flood, coastal erosion, rise of the sea level).

    In the context of the GPSC programme the city has the objective to “integrate climate risks in urban

    planning and management and will focus on urban planning and management, capacity building

    through the development of integrated climate resilience solutions and strengthening the urban

    national policy framework to promote cities’ sustainability at the national level” (GPSC website,

    2018). The project also aims at developing a sustainable cities master plan.

    The main local stakeholder for the city of Dakar is the Directorate of Urbanism and Architecture of the

    Senegalese Ministry of Urban Renewal, Habitat and Living Environment. Other stakeholders include

    mostly other Senegalese governmental agencies.

    The Directorate of Urbanism and architecture, that works in close collaboration with the Ministry of

    Territorial Planning, will use the EO4SD products to develop more accurate policy mechanisms.

    Furthermore, the project will provide a regional view of the environmental dynamics, that will be used

    by the Directorate of Urbanism and Architecture’s team.

    2.2 Service Area Specification

    So far, no internationally accepted definition for the term “Urban Area” and the related Core and Peri-

    Urban areas exists. Different initiatives are currently trying to address a standardised approach for

    defining the terms “Urban Area”. During discussions with the GPSC Co-ordinator it was considered

    important to use a uniform definition for the GPSC cities in order for the cities to exchange

    information and share products/experiences and conduct potential comparative studies.

    In this context, it was decided to use an international approach for the demarcation of the Area of

    Interest (AOI) for mapping the GPSC cities in terms of Core Urban area and Peri-Urban area. Thus,

    the approach is based on the European Union’s Directorate-General for Regional and Urban Policy

    (DG REGIO) method and the definitions are described in the Regional Working Paper 2014 from the

    European Commission on “A harmonised definition of cities and rural areas: the new degree of

    urbanisation” (European Commission, 2014). Following the naming of the DG Regio approach, the

    Urban Core is named as “High Density Core” and the Peri-Urban area is termed as “Urban Cluster”.

    Within the DG REGIO approach, the High Density Core area is defined as contiguous grid cells of 1

    km2 with a density of at least 1 500 inhabitants per km2 and a minimum population of 50 000. The

    Urban Cluster is defined as clusters of contiguous grid cells of 1 km2 with a density of at least 300

    inhabitants per km2 and a minimum population of 5 000.

    The DG REGIO methodology used in the EO4SD-Urban project was slightly adjusted to Non-

    European countries. For the first three GPSC cities (namely Bhopal, Vijayawada and Saint-Louis)

    produced within the project the Global Human Settlement Population (GHSP) grid with a spatial

    resolution of 1 km were used for the classification into “High Density Core” and ”Urban Cluster”. The

    raster dataset is available for the years 1975, 1990, 2000, 2015. This dataset depicts the distribution

    and density of population, expressed as the number of people per cell. The data can be downloaded

    under following link http://data.jrc.ec.europa.eu/dataset/jrc-ghsl-ghs_pop_gpw4_globe_r2015a.

    http://data.jrc.ec.europa.eu/dataset/jrc-ghsl-ghs_pop_gpw4_globe_r2015a

  • Earth Observation for Sustainable Doc. No.: City-Operations Report

    Development – Urban Project Issue/Rev-No.: 1.1

    EO4SD-Urban Dakar City Operations Report Page 3

    In 2019, a higher resolution population layer (spatial resolution of 10 m) produced by the German

    Aerospace Centre (DLR) became available. The AOIs for the remaining GPSC cities (namely Melaka,

    Abidjan, Dakar and Campeche) were produced based on the DLR population layer.

    The High Density Core AOI for a city is created by merging the contiguous grid cells of 1 km2 with a

    density of at least 1500 inhabitants per km2 and a minimum population of 50 000. In the definition of

    the High Density Core the contiguity is only allowed via a vertical or horizontal connection. In a next

    step, gaps are filled. Due to the coarse resolution of the population grid cells additional grid cells were

    in a last step added for under estimated settlement areas. The same was done for over estimations, here

    grid cells were removed. The GHSP layer can be directly used for the calculation, while the DLR

    population has to be aggregated to a resolution of 1 km2 before being used for the AOI definition. In

    this aggregation step, each output cell contains the sum of the input cells that are encompassed by the

    extent of that new cell.

    The Urban Cluster is created very similar to the High Density Core. Continuous grid cells of 1 km2

    with a density of at least 300 inhabitants per km2 and a minimum population of 5 000 are merged

    together to form the Urban Cluster. The contiguity within the Urban Cluster can also be diagonal.

    After gaps are filled, areas, which were over or under estimated by the population grid were removed

    or added to the AOI. The GHSP layer was directly used, the DLR population layer had to undergo an

    aggregation step in order to reduce the spatial resolution to 1 km2.

    For Bhopal and Vijayawada a buffer of 1 km was calculated around the High Density Core AOI and

    the Urban Cluster AOI to smoothen the border of the AOIs.

    In all remaining GPSC cities, the border was not smoothed, but when the population grid was under or

    over estimating the real settlement extent, grid cells were added or removed.

    In some cases, the city counterparts requested that the AOIs for the High Density Core and the Urban

    Cluster follow the municipal or administrative boundary of the city. In this case, the

    municipal/administrative boundary was used, but enlarged in areas where the AOI created according

    to the adjusted DG Regio approach was bigger. This adjustment of the DG Regio AOI was done for

    Melaka, Abidjan, Dakar and Campeche. These further adjusted DG Regio AOIs are in the following

    report named as Core City Area (see Figure 1a) and Larger Urban Area (see Figure 1b).

    A more detailed description on how the AOIs are calculated is provided in Annex 1.

    The AOIs were presented in a power point and sent to the Users for verification. Figure 1 shows the

    resulting AOIs after combining the DG Regio AOIs with the municipal/administrative boundaries of

    the cities.

  • Earth Observation for Sustainable Doc. No.: City-Operations Report

    Development – Urban Project Issue/Rev-No.: 1.1

    EO4SD-Urban Dakar City Operations Report Page 4

    Figure 1: Illustration of Core City and Larger Urban Areas of Dakar.

    The Core City has an area of 422 km2 and the Larger Urban has an area of 823 km2.

  • Earth Observation for Sustainable Doc. No.: City-Operations Report

    Development – Urban Project Issue/Rev-No.: 1.1

    EO4SD-Urban Dakar City Operations Report Page 5

    2.3 Product List and Product Specifications

    During the discussions related to the AOIs the potential geo-spatial products that could be provided for

    the Cities were also reviewed with the WB Team and Users. It was noted that the Baseline Land

    Use/Land Cover (LU/LC) products (for the Core and Peri-Urban areas) were a standard product that

    would be provided for all Cities as it is required for the derived products. In the case of Dakar, the full

    list of products for both the Core and Peri-Urban areas is as follows:

    • Settlement Extent & Change (producer: DLR)

    • Percentage Impervious Surface & Change (producer: DLR)

    • Core City and Larger Urban Land Use / Land Cover (LU/LC) & Change (producer: SIRS)

    • Urban Green Areas & Change (producer: NEO)

    • Flood Hazard & Risk Assessment (producer: JR)

    The first two products have been generated by the German Aerospace Agency (DLR) over the full

    metropolitan area for four reference years: 2000 - 2005 - 2010 - 2015.

    Two time slots were used to provide historic and recent information regarding LU/LC for Dakar, 2006

    and 2018 over the Core City and Larger Urban Areas. The last section of the Report is fully dedicated

    to the Flood Hazard & Risk Assessment study.

    2.4 Land Use/Land Cover Nomenclature

    A pre-cursor to starting production was the establishment with the stakeholders on the relevant Land

    Use/Land Cover (LU/LC) nomenclature as well as class definitions. The approach taken was to use a

    standard remote sensing based LU/LC nomenclature i.e. the European Urban Atlas Nomenclature

    (European Union, 2011) and adapt it to the User’s LU requirements. Thus, the remote-sensing based

    LU/LC classes in the urban context can be grouped into five Level 1 classes, which are Artificial

    Surfaces, Natural/ Semi Natural Areas, Agricultural Areas, Wetlands, and Water. These classes can

    then be sub-divided into several different more detailed classes such that the dis-aggregation can be

    down to Level 2-4. This hierarchical classification system is often used in operational Urban mapping

    programmes and is the basis for example of the European Commission’s Urban Atlas programme

    which provides pan-European comparable LU/LC data with regular updates. A depiction of the way

    the levels and classes in the Urban Atlas programme are structured is presented as follows:

  • Earth Observation for Sustainable Doc. No.: City-Operations Report

    Development – Urban Project Issue/Rev-No.: 1.1

    EO4SD-Urban Dakar City Operations Report Page 6

    Level I Artificial surfaces

    - Level II: Urban Fabric

    Level III

    • Continuous Urban Fabric (Sealing Layer-S.L. > 80%)

    • Discontinuous Urban Fabric (S.L. 10% - 80%)

    Level IV

    1) Discontinuous Dense Urban Fabric (S.L. 50% - 80%)

    2) Discontinuous Medium Density Urban Fabric (S.L. 30% - 50%)

    3) Discontinuous Low Density Urban Fabric (S.L. 10% - 30%)

    4) Discontinuous Very Low Density Urban Fabric (S.L. < 10%)

    - Level II: Industrial, Commercial, Public, Military, Private Units and Transport

    Level III

    • Industrial, Commercial, Public, Military and Private Units

    • Transport Infrastructure

    Level IV

    5) Fast Transit Roads

    6) Other Roads

    7) Railway

    • Port and associated land

    • Airport and associated land

    - Level II: Mine, Dump and Construction Sites

    Level III

    • Mineral Extraction and Dump Sites

    • Construction Sites

    • Land Without Current Use

    - Level II: Artificial Non-Agricultural Vegetated Areas

    Level III

    • Green Urban Areas

    • Sports and Leisure Facilities

    (Reference: European Union, 2011)

    It should be noted that in the current project, the Level 1 classes were used as the basis for

    classification of the Urban Cluster areas using the High Resolution (HR) data such as Landsat or

    Sentinel. However, for the High Density Core areas using the Very High Resolution (VHR) data it was

    possible to go down to Level III and IV.

    The different levels, classes and sub-classes from the remote sensing based urban classification, were

    harmonised within the GPSC cities. The following tables give the nomenclature for the High Density Core

    and the Urban Cluster region (see Table 1 and Table 2).

  • Earth Observation for Sustainable Doc. No.: City-Operations Report

    Development – Urban Project Issue/Rev-No.: 1.1

    EO4SD-Urban Dakar City Operations Report Page 7

    Table 1: LU/LC Nomenclature for GPSC Cities (Core City AOI).

    Actual and Historic Nomenclature Core City Area

    Level I Level II Level III Level IV

    1000 Artificial

    Surfaces

    1100 Residential 1110 Continuous Urban Fabric

    (80 - 100 % Sealed)

    1120 Discontinuous Urban

    Fabric

    1121 Discontinuous dense urban fabric (50 - 80 % Sealed)

    1122 Discontinuous medium density urban fabric (30 - 50 % Sealed)

    1123 Discontinuous low density urban fabric (10 - 30 % Sealed)

    1124 Discontinuous very low density urban fabric (0 - 10 % Sealed)

    1200 Industrial,

    Commercial, Public,

    Military, Private Units

    and Transport

    1210 Industrial, Commercial,

    Public, Military and Private

    Units

    1220 Transport Infrastructure 1221 Arterial Roads

    1222 Collector Roads

    1223 Railway

    1230 Port Area

    1240 Airport

    1300 Mine, Dump and

    Construction Sites

    1310 Mineral Extraction and

    Dump Sites

    1330 Construction Sites

    1340 Land Without Current

    Use

    1400 Artificial Non-

    Agricultural Vegetated

    Areas

    1410 Green Urban Areas

    1420 Sports and Leisure

    Facilities

    2000 Agricultural

    Area

    3000 Natural and

    Semi-natural Areas

    3100 Forest and

    Shrublands

    3200 Natural Areas

    (Grassland)

    3300 Bare Soil

    4000 Wetlands

    5000 Water 5100 Inland Water

    5200 Marine Water

    Table 2: LU/LC Nomenclature for GPSC Cities (Larger Urban AOI).

    Actual and Historic Nomenclature Larger Urban Area

    Level I Level II Level III Level IV

    1000 Artificial

    Surfaces

    2000 Agricultural

    Area

    3000 Natural and

    Semi-natural Areas

    3100 Forest and

    Shrublands

    3200 Natural Areas

    (Grassland)

    3300 Bare Soil

    4000 Wetlands

    5000 Water 5100 Inland Water

    5200 Marine Water

    It is important to note that the possibility to classify at Level IV is highly dependent on the availability

    of reliable reference datasets from the City or sources such as Google Earth. This aspect is further

    discussed in Chapter 3.

    Especially regarding the road hierarchy used in the classification at Level IV, the international road

    classification standards have been followed; this is for example defined by the European Commission

    (https://ec.europa.eu/transport/road_safety/specialist/-

    knowledge/road/designing_for_road_function/road_classification_en).

  • Earth Observation for Sustainable Doc. No.: City-Operations Report

    Development – Urban Project Issue/Rev-No.: 1.1

    EO4SD-Urban Dakar City Operations Report Page 8

    Roads are divided into three groups: arterial or through traffic flow routes (in our case Arterial

    Roads), distributor roads (in our case Collector Roads), and access roads (or Local Roads). The three

    road types are defined as follows:

    Arterial Roads:

    Roads with a flow function allow efficient throughput of (long distance) motorized traffic. All

    motorways and express roads as well as some urban ring roads have a flow function. The number of

    access and exit points is limited. (https://ec.europa.eu/transport/road_safety/specialist/knowledge/-

    road/designing_for_road_function/road_classification_en)

    Collector Roads:

    Roads with an area distributor function allow entering and leaving residential areas, recreational areas,

    industrial zones, and rural settlements with scattered destinations. Junctions are for traffic exchange

    (allowing changes in direction etc.); road sections between junctions should facilitate traffic in

    flowing.

    (https://ec.europa.eu/transport/road_safety/specialist/knowledge/road/designing_for_road_function/roa

    d_classification_en)

    Local Roads:

    Roads with an access function allow actual access to properties alongside a road or street. Both

    junctions and the road sections between them are for traffic exchange. (https://ec.europa.eu/transport/-

    road_safety/specialist/knowledge/road/designing_for_road_function/road_classification_en).

  • Earth Observation for Sustainable Doc. No.: City-Operations Report

    Development – Urban Project Issue/Rev-No.: 1.1

    EO4SD-Urban Dakar City Operations Report Page 9

    2.5 World Settlement Extent

    Reliably outlining settlements is of high importance since an accurate characterization of their extent

    is fundamental for accurately estimating, among others, the population distribution, the use of

    resources (e.g. soil, energy, water, and materials), infrastructure and transport needs, socioeconomic

    development, human health and food security. Moreover, monitoring the change in the extent of

    settlements over time is of great support for properly modelling the temporal evolution of urbanization

    and thus, better estimating future trends and implementing suitable planning strategies.

    At present, no standard exists for defining settlements and worldwide almost each country applies its

    own definition either based on population, administrative or geometrical criteria. The German Space

    Agency (DLR) was responsible for the provision of the “Settlement Extent” product; when generating

    the settlement extent maps from HR imagery, pixels are labelled as settlement if they intersect any

    building, lot or – just within urbanized areas – roads and paved surface where we define:

    • building as any structure having a roof supported by columns or walls and intended for the

    shelter, housing, or enclosure of any individual, animal, process, equipment, goods, or

    materials of any kind;

    • lot as the area contained within an enclosure (wall, fence, hedge) surrounding a building or a

    group of buildings. In cases where there are many concentric enclosures around a building, the

    lot is considered to stop at the inner most enclosure;

    • road as any long, narrow stretch with a smoothed or paved surface, made for traveling by

    motor vehicle, carriage, etc., between two or more points;

    • paved surface as any level horizontal surface covered with paving material (i.e., asphalt,

    concrete, concrete pavers, or bricks but excluding gravel, crushed rock, and similar materials).

    Instead, pixels not satisfying this condition are marked as non-settlement.

    The settlement extent product is a binary mask outlining - in the given Area of Interest (AOI) –

    settlements in contrast to all other land-cover classes merged together into a single information class.

    The settlement class and the non-settlement class are associated with values “255” and “0”,

    respectively.

  • Earth Observation for Sustainable Doc. No.: City-Operations Report

    Development – Urban Project Issue/Rev-No.: 1.1

    EO4SD-Urban Dakar City Operations Report Page 10

    2.6 Percentage Impervious Surface

    Settlement growth is associated not only to the construction of new buildings, but – more in general –

    to a consistent increase of all the impervious surfaces (hence also including roads, parking lots,

    squares, pavement, etc.), which do not allow water to penetrate, forcing it to run off. To effectively

    map the percentage impervious surface (PIS) is then of high importance being it related to the risk of

    urban floods, the urban heat island phenomenon as well as the reduction of ecological productivity.

    Moreover, monitoring the change in the PIS over time is of great support for understanding, together

    with information about the spatiotemporal settlement extent evolution, also more details about the type

    of urbanization occurred (e.g. if areas with sparse buildings have been replaced by highly impervious

    densely built-up areas or vice-versa).

    In the framework of the EO4SD-Urban project, one pixel in the generated PIS maps is associated with

    the estimated percentage of the corresponding surface at the ground covered by buildings or paved

    surfaces, are defined as:

    • building as any structure having a roof supported by columns or walls and intended for the

    shelter, housing, or enclosure of any individual, animal, process, equipment, goods, or materials

    of any kind;

    • paved surface as any level horizontal surface covered with paving material (i.e. asphalt, concrete,

    concrete pavers, or bricks but excluding gravel, crushed rock, and similar materials).

    The product provides for each pixel in the considered AOI the estimated PIS. Specifically, values are

    integer and range from 0 (no impervious surface in the given pixel) to 100 (completely impervious

    surface in the given pixel) with step 5.

    2.7 Urban Green Areas Nomenclature

    Developing cities in a sustainable way implies to preserve and promote green areas also and especially

    within the urban extent. Green areas refer to any surfaces covered by vegetation (grass, bushes, trees).

    Table 1: Nomenclature used for the mapping and identification of Urban Green Areas.

    Single date

    Code 0 Non-urban green area

    Code 1 Urban green area

    Code 255 Non-urban areas. All areas that do not fall in “Artificial Surfaces” Level 1 class of

    the Land Use Land Cover product (See Table 1).

    Change product

    Code 0 Non-urban green area. No vegetated surfaces occurring on “Artificial Surfaces”,

    Level I, at both points in time.

    Code 1 Permanent urban green area. Vegetated surfaces in historic and recent year.

    Code 2 Loss of urban green area. Vegetated areas in historic year, which changed to non-

    vegetated areas in recent year.

    Code 3 New urban green area. Non-vegetated surfaces in historic year with vegetation

    cover in recent year.

    Code 255 Non-Urban Areas. All areas that do not fall in “Artificial Surfaces” Level 1 class of

    the Land Use Land Cover product.

    2.8 Terms of Access

    The Dissemination of the digital data and the Report was undertaken via FTP.

  • Earth Observation for Sustainable Doc. No.: City-Operations Report

    Development – Urban Project Issue/Rev-No.: 1.1

    EO4SD-Urban Dakar City Operations Report Page 11

    3 Service Operations

    The following Sections present all steps of the service operations including the necessary input data,

    the processing methods, the accuracy assessment and the Quality Control procedures. Methods are

    presented in a top-level and standardised manner for all the EO4SD-Urban City Reports.

    3.1 Source Data

    This Section presents a summary of the remote sensing and ancillary datasets that were used. Different

    types of data from several data providers have been acquired. A complete list of source data as well as

    a quality assessment is provided in Annex 3.

    High Resolution Optical EO Data

    The major data sources for the current and historic mapping of LULC for Larger Urban Area,

    Settlement Extent and PIS products were Landsat and Sentinel-2 data which were accessible and

    downloadable free of charge.

    • Landsat 7: As a source of historical data four scenes of Landsat TM 7 from the 14th of

    January 2006 to 28th of March have been acquired which covers the whole area of interest.

    • Sentinel-2: The most recent data coverage comprises one Sentinel-2 data set from the 27th of

    February 2018. The data was downloaded and processed at Level 1C.

    Very High Resolution Optical EO Data

    The VHR data for the Core Urban Area mapping had to be acquired and purchased through

    commercial EO Data Providers such as Airbus Defence and European Space Imaging.

    It has to be noted that under the current collaboration project the VHR EO data had to be purchased

    under mono-license agreements between GAF AG and the EO Data Providers. If EO data would have

    to be distributed to other stakeholders, then further licences for multiple users would have to be

    purchased.

    The following VHR sensor data have been acquired to cover the AOI:

    • Quickbird-2:

    o 4 scenes from the 07 November 2005 to the 21th of December 2005 covering 99.3% of

    the Core Urban AOI

    • Pléiades 1-A & Pléiades 1-B:

    o 3 scenes from the 1st of March 2018 to the 3rd of March 2018 covering 99.9% of the

    Core Urban AOI

    Detailed lists of the used EO data as well as their quality is documented in the attached Quality

    Control Sheets in Annex 3.

    Ancillary Data

    Open Street Map (OSM) data: OSM data is freely available and generated by volunteers across the

    globe. The so-called crowd sourced data is not always complete but has for the most parts of the world

    valuable spatial information. Data was downloaded to complement the Transport Network layer and

    further enhanced. The spatial location of the OSM based streets was used a geospatial reference.

    Detailed lists of the used EA and ancillary data as well as their quality is documented in the attached

    Quality Control Sheets in Annex 3.

  • Earth Observation for Sustainable Doc. No.: City-Operations Report

    Development – Urban Project Issue/Rev-No.: 1.1

    EO4SD-Urban Dakar City Operations Report Page 12

    3.2 Processing Methods

    Data processing starts at an initial stage with quality checks and verification of all incoming data. This

    assessment is performed in order to guarantee the correctness of data before geometric or radiometric

    pre-processing is continued. These checks follow defined procedures in order to detect anomalies,

    artefacts and inconsistencies. Furthermore, all image and statistical data were visualized and

    interpreted by operators.

    The main techniques and standards used for data analysis, processing and modelling for each product

    are described in Annex 2.

    3.3 Accuracy Assessment of Map Products

    Data and maps derived from remote sensing contain - like any other map - uncertainties which can be

    caused by many factors. The components, which might have an influence on the quality of the maps

    derived from EO include quality and suitability of satellite data, interoperability of different sensors,

    radiometric and geometric processing, cartographic and thematic standards, and image interpretation

    procedures, post-processing of the map products and finally the availability and quality of reference

    data. However, the accuracy of map products has a major impact on secondary products and its utility

    and therefore an accuracy assessment was considered as a critical component of the entire production

    and products delivery process. The main goal of the thematic accuracy assessment was to guarantee

    the quality of the mapping products with reference to the accuracy thresholds set by the user

    requirements.

    The applied accuracy assessments were based on the use of reference data and applying statistical

    sampling to deduce estimates of error in the classifications. In order to provide an efficient, reliable

    and robust method to implement an accuracy assessment, there are three major components that had to

    be defined: the sampling design, which determines the spatial location of the reference data, the

    response design that describes how the reference data is obtained and an analysis design that defines

    the accuracy estimates. These steps were undertaken in a harmonized manner for the validation of all

    the geo-spatial products.

    3.3.1 Accuracy Assessment of the LU/LC Products

    Sampling Design

    The sampling design specifies the sample size, sample allocation and the reference assessment units

    (i.e. pixels or image blocks). Generally, different sampling schemes can be used in collecting

    accuracy assessment data including: simple random sampling, systematic sampling, stratified

    random sampling, cluster sampling, and stratified systematic unaligned sampling. In the current

    project a single stage stratified random sampling based on the method described by Olofson et al

    (20131) was applied which used the map product as the basis for stratification. This ensured that all

    classes, even very minor ones were included in the sample.

    The sampling design is applied separately for the Core City Area and for the Larger Urban Area

    classification.

    1 Olofsson, P., Foody, G. M., Stehman, S. V., & Woodcock, C. E. (2013). Making better use of accuracy data in

    land change studies: Estimating accuracy and area and quantifying uncertainty using stratified estimation.

    Remote Sensing of Environment, 129, 122–131. doi:10.1016/j.rse.2012.10.031

  • Earth Observation for Sustainable Doc. No.: City-Operations Report

    Development – Urban Project Issue/Rev-No.: 1.1

    EO4SD-Urban Dakar City Operations Report Page 13

    In the complex LU/LC product with many classes, this usually results in a large number of strata

    (one stratum per LU/LC classes), of which some classes cover only very small areas (e.g. sport

    fields, cemeteries) and not being adequately represented in the sampling. In order to achieve a

    representative sampling for the statistical analyses of the mapping accuracy it was decided to extend

    the single stage stratified random sampling. Slightly different approaches were used for the Core

    City and the Larger Urban Area classification.

    The first step is the same for both classifications: the number of required samples is allocated within

    each of the Level I strata (1000 Artificial Surfaces, 2000 Agricultural Area, 3000 Natural and Semi-

    natural Areas, 4000 Wetlands, 5000 Water).

    In the second step, all Level III classes that were not covered by the first sampling were grouped into

    one new stratum for the Core City Area classification. For the Larger Urban Area classification all

    Level II classes that were not covered by the first sampling were grouped into one new stratum.

    Within that stratum the same number of samples was randomly allocated as the Level I strata

    received. To avoid a clustering of point samples within classes and to minimise the effect of spatial

    autocorrelation a minimum distance in between the sample points was set to be 150 m. The final

    sample size for each class can be considered to be as close as possible to the proportion of the area

    covered by each stratum considering that the target was to determine the overall accuracy of the

    entire map.

    The total sample size per stratum was determined by the expected standard error and the estimated

    error rate based on the following formula, which assumes a simple random sampling (i.e. the

    stratification is not considered):

    n = 𝑃∗𝑞

    (𝐸

    𝑧)²

    n = number of samples per strata / map class

    p = expected accuracy

    q = 1 – p

    E = Level of acceptable (allowable) sample error

    Z = z-value (the given level of significance)

    Hence, with an expected accuracy of p = 0.85, a 95% confidence level and an acceptable sampling

    error of 5%, the minimum sample size is 196. A 10% oversampling was applied to compensate for

    stratification inefficiencies and potentially inadequate samples (e.g. in case of cloudy or shady

    reference data). For each Level I strata 215 samples have been randomly allocated. Afterwards, for

    all classes of Level III of the Core City Area classification that did not received samples in the first

    run, additionally 215 samples were randomly drawn across all these classes. A summary of the

    number of sample point for each Core City Area class is given in Table 3.

    The same applies for the Larger Urban Area classification: All Level II classes that did not receive

    samples in the first run, additionally 215 samples were randomly drawn across all these classes. A

    summary of the number of sample point for each Larger Urban Area class is given in

    Table 4.

    The main difference of the sampling design for the two areas is that the resampling is done at Level III

    for the Core City Area and at Level II for the Larger Urban Area.

  • Earth Observation for Sustainable Doc. No.: City-Operations Report

    Development – Urban Project Issue/Rev-No.: 1.1

    EO4SD-Urban Dakar City Operations Report Page 14

    Table 3: Number of sampling points for the Core City Area classes after applied sampling design with

    information on overall land cover by class.

    Class Name Class ID

    No. Of

    Sampling

    Points

    Km2 Coverage

    Continuous Urban Fabric 1110 196 108.2

    Discontinuous Urban Fabric 1120 19 7.1

    Industrial, Commercial, Public,

    Military and Private Units 1210 148 31.4

    Transport Infrastructure 1220 31 7.9

    Port Area 1230 12 1.9

    Airport 1240 24 5.5

    Mineral Extraction and Dump Sites 1310 28 3.5

    Construction Sites 1330 32 4.1

    Land Without Current Use 1340 141 17.3

    Green Urban Areas 1410 20 2.3

    Sports and Leisure Facilities 1420 20 2.6

    Agricultural Area 2000 180 22.3

    Forest and Shrublands 3100 40 5.0

    Natural Areas (Grassland) 3200 151 18.7

    Bare Soil 3300 107 13.2

    Wetlands 4000 59 7.3

    Inland Water 5100 27 3.3

    Marine Water 5200 215 160.7

    Total - 1450 422.4

    Table 4: Number of sampling points for the Larger Urban Area classes after applied sampling design with

    information on overall land cover by class.

    Class Name Class

    ID

    No. Of

    Sampling

    Points

    Km2 Coverage

    Artificial Surfaces 1000 215 259.9

    Agriculture 2000 215 243.6

    Forest and Shrublands 3100 61 15.5

    Natural Areas 3200 215 54.8

    Bare Soil 3300 75 29.1

    Wetlands 4000 31 10.3

    Inland Water 5100 27 5.8

    Marine Water 5200 215 204.4

    Total - 1054 823.4

  • Earth Observation for Sustainable Doc. No.: City-Operations Report

    Development – Urban Project Issue/Rev-No.: 1.1

    EO4SD-Urban Dakar City Operations Report Page 15

    Response Design

    The response design determines the reference information for comparing the map labels to the

    reference labels. Collecting reference data on the ground by means of intensive fieldwork is both

    costly and time-consuming and is in most projects not feasible. The most cost-effective reference data

    sources are VHR satellite data with 0.5 m to 1 m spatial resolution. Czaplewski (2003)2 indicated that

    visual interpretation of EO data is acceptable if the spatial resolution of EO data is sufficiently better

    compared to the thematic classification system. However, if there are no EO data with better spatial

    resolution available, the assessment results need to be checked against the imagery used in the

    production process.

    The calculated number of necessary sampling points for each mapping category was randomly

    distributed among the strata and overlaid onto the two LULC mapping products. The following two

    Figures (see Figure 2 and Figure 3) are showing the mapping result with the overlaid sample points.

    Figure 2: Mapping result of the Core City Area of Dakar of the year 2018 overlaid with randomly

    distributed sample points used for accuracy assessment.

    2 Czaplewski, R. L. (2003). Chapter 5: accuracy assessment of maps of forest condition: statistical design and

    methodological considerations, pp. 115–140. In Michael A.Wulder, & Steven E. Franklin (Eds.), Remote

    sensing of forest environments: concepts and case studies. Boston: Kluwer Academic Publishers (515 pp.).

  • Earth Observation for Sustainable Doc. No.: City-Operations Report

    Development – Urban Project Issue/Rev-No.: 1.1

    EO4SD-Urban Dakar City Operations Report Page 16

    Figure 3: Mapping result of the Larger Urban Area of Dakar of the year 2018 overlaid with randomly

    distributed sample points used for accuracy assessment.

    In this way a reference information could be extracted for each sample point by visual interpretation of

    the VHR data for all mapped classes. The size of the area to be observed had to be related to the

    Minimum Mapping Unit (MMU) of the map product to be assessed. The reference information of each

    sampling point was compared with the mapping results and the numbers of correctly and not-correctly

    classified observations were recorded for each class. From this information the specific error matrices

    and statistics were computed (see next Section).

    Analysis

    Each class usually has errors of both omission and commission, and in most situations, these errors for

    a class are not equal. In order to calculate these errors as well as the uncertainties (confidence

    intervals) for the area of each class a statistically sound accuracy assessment was implemented.

    The confusion matrix is a common and effective way to represent quantitative errors in a categorical

    map, especially for maps derived from remote sensing data. The matrices for each assessment epoch

    were generated by comparing the “reference” information of the samples with their corresponding

    classes on the map. The Reference represented the “truth”, while the Map provided the data obtained

    from the map result. Thematic accuracy for each class and overall accuracy is then presented in error

    matrices. Unequal sampling intensity resulting from the random sampling approach was accounted for

    by applying a weight factor (p) to each sample unit based on the ratio between the number of samples

    and the size of the stratum considered3:

    3 Selkowitz, D. J., & Stehman, S. V. (2011). Thematic accuracy of the National Land Cover Database (NLCD)

    2001 land cover for Alaska. Remote Sensing of Environment, 115(6), 1401–1407.

    doi:10.1016/j.rse.2011.01.020.

  • Earth Observation for Sustainable Doc. No.: City-Operations Report

    Development – Urban Project Issue/Rev-No.: 1.1

    EO4SD-Urban Dakar City Operations Report Page 17

    �̂�𝑖𝑗 = (1

    𝑀) ∑

    1

    𝜋𝑢ℎ∗

    𝑥∈(𝑖,𝑗)

    Where i and j are the columns and rows in the matrix, M is the total number of possible units

    (population) and π is the sampling intensity for a given sample unit u in stratum h.

    Overall accuracy and User and producer accuracy were computed for all thematic classes and 95%

    confidence intervals were calculated for each accuracy metric.

    The standard error of the error rate was calculated as follows: 𝜎ℎ = √𝑝ℎ(1−𝑝ℎ)

    𝑛ℎ where nh is the sample

    size for stratum h and ph is the expected error rate. The standard error was calculated for each stratum

    and an overall standard error was calculated based on the following formula:

    𝜎 = √∑ 𝑤ℎ2. 𝜎ℎ

    2

    In which 𝑤ℎ is the proportion of the total area covered by each stratum. The 95% Confidence Interval

    (CI) is +/- 1.96*𝜎.

    Results

    The confusion matrices are provided within the Annex 3 and show the mapping error for each relevant

    class. For each class the number of samples which are correctly and not correctly classified are listed,

    this allows the calculation of the user and producer accuracies for each class as well as the confidence

    interval at 95% confidence levels based on the formulae above.

    The Land Use/Land Cover product for Dakar in 2018 in Core City Area has an overall mapping

    accuracy of 98.94% with a CI ranging from 98.41% to 99.47% at a 95% CI.

    For the Larger Urban Area, the overall accuracy is 89.23% with a CI ranging from 87.36% to

    91.1% at a 95% CI. The specific class accuracies are given in Annex 3.

  • Earth Observation for Sustainable Doc. No.: City-Operations Report

    Development – Urban Project Issue/Rev-No.: 1.1

    EO4SD-Urban Dakar City Operations Report Page 18

    3.3.2 Accuracy Assessment of the World Settlement Extent Product

    In the following, the strategy designed for validating the World Settlement Extent (WSE) or World

    Settlement Footprint (WSF) 2015, i.e. a global settlement extent layer obtained as a mosaic of ~18.000

    tiles of 1x1 degree size where the same technique employed in the EO4SD-Urban project is presented.

    In particular, specific details are given for all protocols adopted for each of the accuracy assessment

    components, namely response design, sampling design, and analysis; final results are discussed

    afterwards. In the light of the quality and amount of validation points considered, it can be reasonably

    assumed that the corresponding quality assessment figures are also representative for any settlement

    extent map generated in the framework of EO4SD-Urban.

    Response Design

    The response design encompasses all steps of the protocol that lead to a decision regarding agreement

    of the reference and map classifications. The four major features of the response design are the source

    of information used to determine the source of reference data, the spatial unit, the labelling protocol

    for the reference classification, and a definition of agreement.

    • Source of Reference Data: Google Earth (GE) satellite/aerial VHR imagery has been used given

    its free access and the availability for all the project test sites in the period 2014-2015. In

    particular, GE automatically displays the latest available data, but it allows to browse in time over

    all past historical images. The spatial resolution varies depending on the specific data source; in

    the case of SPOT imagery it is ~1.5m, for Digital Globe's WorldView-1/2 series, GeoEye-1, and

    Airbus' Pleiades it is in the order of ~0.5m resolution, whereas for airborne data (mostly available

    for North America, Europe and Japan) it is about 0.15m.

    • Spatial Assessment Unit: A 3x3 block spatial assessment unit composed of 9 cells of 10x10m

    size has been used. Specifically, this choice is justified one the one hand by the fact that input

    data with different spatial resolutions have been used to generate the WSF2015 (i.e. 30m Landsat-

    8 and 10m S1). On the other hand, GE imagery exhibited in some cases a mis-registration error of

    the order of 10-15m, hence using a 3x3 block allows defining an agreement e.g. based on

    statistics computed over 9 pixels, thus reducing the impact of such shift.

    • Reference Labelling Protocol: For each spatial assessment block any cell is finally labelled as

    settlement if it intersects any building, lot or – just within settlements – roads and paved surface.

    Instead, pixels not satisfying this condition are marked as non-settlement.

    • Definition of Agreement: Given the classification and the reference labels derived as described

    above, three different agreement criteria have been defined:

    8) for each pixel, positive agreement occurs only for matching labels between the

    classification and the reference;

    9) for each block, a majority rule is applied over the corresponding 9 pixels of both the

    classification and the reference; if the final labels match, then the agreement is positive;

    10) for the classification a majority rule is applied over each assessment block, while for the

    reference each block is labelled as “settlement” only in the case it contains at least one

    pixel marked as “settlement”; if the final labels match, then the agreement is positive.

    Crowd-sourcing was performed internally at Google. In particular, by means of an ad-hoc tool,

    operators have been iteratively prompted a given cell on top of the available Google Earth reference

    VHR scene closest in time to the year 2015 and given the possibility of assigning to each cell a label

    among: “building”, “lot”, “road/paved surface” and “other”. For training the operators, a

    representative set of 100 reference grids was prepared in collaboration between Google and DLR.

  • Earth Observation for Sustainable Doc. No.: City-Operations Report

    Development – Urban Project Issue/Rev-No.: 1.1

    EO4SD-Urban Dakar City Operations Report Page 19

    Sampling Design

    The stratified random sampling design has been applied since it satisfies the basic accuracy assessment

    objectives and most of the desirable design criteria. In particular, stratified random sampling is a

    probability sampling design and it is one of the easier to implement; indeed, it involves first the

    division of the population into strata within which random sampling is performed afterwards. To

    include a representative population of settlement patterns, 50 out of the ~18.000 tiles of 1x1 degree

    size considered in the generation of the WSF2015 have been selected based on the ratio between the

    number of estimated settlements (i.e. disjoint clusters of pixels categorized as settlement in the

    WSF2015) and their area. In particular, the i-th selected tile has been chosen randomly among those

    whose ratio belongs to the interval ]𝑃2(𝑖−1); 𝑃2𝑖], 𝑖 ∈ [1; 50] ⊂ ℕ (where 𝑃𝑥 denotes the x-th percentile

    of the ratio).

    Table 5: Accuracies exhibited by the WSF2015 according to the three considered agreement criteria for

    different definitions of settlement.

    Settlement = Accuracy

    Measure

    Agreement Criterion

    1 2 3

    buildings

    OA% 86.96 87.86 91.15

    AA% 88.57 90.35 88.91

    Kappa 0.6071 0.6369 0.7658

    UANS% - UAS% 98.11 54.69 98.73 56.76 94.84 80.58

    PANS% - PAS% 86.24 90.90 86.72 93.98 93.32 84.51

    buildings + lots

    OA 88.08 88.94 91.26

    AA% 88.64 90.19 88.71

    Kappa 0.6510 0.6784 0.7716

    UANS% - UAS% 97.54 60.71 98.13 62.66 94.29 82.62

    PANS% - PAS% 87.79 89.49 88.26 92.12 93.95 83.48

    buildings + lots

    + roads / paved

    surface

    OA 88.77 90.09 88.51

    AA% 86.34 88.28 84.27

    Kappa 0.6938 0.7317 0.7219

    UANS% - UAS% 94.49 72.20 95.35 75.06 88.13 89.60

    PANS% - PAS% 90.78 81.91 91.62 84.94 96.04 72.51

    As the settlement class covers a sensibly small proportion of area compared to the merger of all other

    non-settlement classes (~1% of Earth’s emerged surface), an equal allocation reduces the standard

    error of its class-specific accuracy. Moreover, such an approach allows to best address user’s accuracy

    estimation, which corresponds to the map “reliability” and is indicative of the probability that a pixel

    classified on the map actually represents the corresponding category on the ground. Accordingly, in

    this framework for each of the 50 selected tiles we randomly extracted 1000 settlement and 1000 non-

    settlement samples from the WSF2015 and used these as centre cells of the 3x3 reference block

    assessment units to label by photointerpretation. Such a strategy resulted in an overall amount of

    (1000 + 1000) × 9 × 50 = 900.000 cells labelled by the crowd.

    Analysis

    As measures for assessing the accuracy of the settlement extent maps, we considered:

    • the percentage overall accuracy OA%;

    • the Kappa coefficient;

    • the percentage producer’s (PAS%, PANS%) and user’s (UAS%, UANS%) accuracies for both the

    settlement and non-settlement class;

    • the percentage average accuracy AA% (i.e., the average between PAS% and PANS%).

  • Earth Observation for Sustainable Doc. No.: City-Operations Report

    Development – Urban Project Issue/Rev-No.: 1.1

    EO4SD-Urban Dakar City Operations Report Page 20

    Results

    Table 5 reports the accuracies exhibited by the WSF2015 according to the three considered agreement

    criteria for different definitions of settlement; specifically, we considered as “settlement” all areas

    covered by: i) buildings; ii) buildings or building lots; or iii) buildings, building lots or roads / paved

    surfaces. As one can notice, accuracies are always particularly high, thus confirming the effectiveness

    of the employed approach and the reliability of the final settlement extent maps. The best

    performances in terms of kappa are obtained when considering settlements as composed by buildings,

    building lots and roads / paved surfaces for criteria 1 and 2 (i.e., 0.6938 and 0.7317, respectively) and

    by buildings and building lots for criteria 3 (0.7716); the OA% follows a similar trend. This is in line

    with the adopted settlement definition. Moreover, agreement criteria 3 results in accuracies

    particularly high with respect to criteria 1 and 2 when considering as settlement just buildings or the

    combination of buildings and lots. This can be explained by the fact that when the detection is mainly

    driven by Landsat data then the whole 3x3 assessment unit tends to be labelled as settlement if a

    building or a lot intersect the corresponding 30m resolution pixel.

    3.3.3 Accuracy Assessment of the Percentage Impervious Surface Product

    In the following section, the strategy designed for validating the PIS product is presented; specifically,

    details are given for all protocols adopted for each of the accuracy assessment components, namely

    response design, sampling design, and analysis. Results are discussed afterwards.

    Response Design

    The response design encompasses all steps of the protocol that lead to a decision regarding agreement

    of the reference and map classifications. The four major features of the response design are the source

    of information used to determine the source of reference data, the spatial unit, the labelling protocol

    for the reference classification, and a definition of agreement.

    • Source of Reference Data: Cloud-free VHR multi-spectral imagery (Visible + Near Infrared)

    acquired at 2m spatial resolution (or higher) covering a portion of the AOI for which the Landsat-

    based PIS product has been generated;

    • Spatial Assessment Unit: A 30x30m size unit has been chosen according to the spatial resolution

    of the Landsat imagery employed to generate the PIS product;

    • Reference Labelling Protocol: As a first step, the NDVI is computed for each VHR scene

    followed by a manual identification of the most suitable threshold that allows to exclude all the

    vegetated areas (i.e. non-impervious). Then, the resulting mask is refined by extensive

    photointerpretation.

    • Definition of Agreement: The above-mentioned masks are aggregated at 30m spatial resolution

    and compared per-pixel with the resulting VHR-based reference PIS to the corresponding portion

    of the Landsat-based PIS product.

    Sampling Design

    The entirety of pixels covered by the available VHR imagery over the given AOI is employed for

    assessing the quality of the Landsat-based PIS product.

  • Earth Observation for Sustainable Doc. No.: City-Operations Report

    Development – Urban Project Issue/Rev-No.: 1.1

    EO4SD-Urban Dakar City Operations Report Page 21

    Analysis

    As measures for assessing the accuracy of the PIS maps, following indices are computed:

    • the Pearson’s Correlation coefficient: it measures the strength of the linear relationship between

    two variables and it is defined as the covariance of the two variables divided by the product of

    their standard deviations; in particular, it is largely employed in the literature for validating the

    output of regression models;

    • The Mean Error (ME): it is calculated as the difference between the estimated value (i.e., the

    Landsat-based PIS) and the reference value (i.e., the VHR-based reference PIS) averaged over all

    the pixels of the image;

    • The Mean Absolute Error (MAE): it is calculated as the absolute difference between the estimated

    value (i.e., the Landsat-based PIS) and the reference value (i.e., the VHR-based reference)

    averaged over all the pixels of the image.

    Results

    To assess the effectiveness of the method developed to generate the PIS maps, its performances over 5

    test sites is analysed (i.e. Antwerp, Helsinki, London, Madrid and Milan) by means of WorldView-2

    (WV2) scenes acquired in 2013-2014 at 2m spatial resolution. In particular, given the spatial detail

    offered by WV2 imagery, it was possible to delineate with a very high degree of confidence all the

    buildings and other impervious surfaces included in the different investigated areas. Details about

    acquisition date and size are reported in Table 6, along with the overall number of final 30x30m

    validation samples derived for the validation exercise. Such a task demanded a lot of manual

    interactions and transferring it to other AOIs would require extensive efforts; however, it can be

    reasonably assumed that the final quality assessment figures (computed on the basis of more than 1.9

    million validation samples) shall be considered representative also for PIS maps generated in the

    framework of EO4SD-Urban. Table 6 reports the quantitative results of the comparison between the

    PIS maps generated using Landsat-7/8 data acquired in 2013-2014 and the WV2-based reference PIS

    maps. In particular, the considered approach allowed to obtain a mean correlation of 0.8271

of 109/109
Earth Observation for Sustainable Development Urban Development Project This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 685761. ESA Ref: AO/1-8346/15/I-NB Doc. No.: City Operations Report Issue/Rev.: 1.1 Date: 19.11.2019 EO4SD-Urban Project: Dakar City Report Partners: Financed by:
Embed Size (px)
Recommended